Hi, I’m Raiyan.

Work, explained clearly.

Six projects across the stack. Open one for the problem, the idea, and what came of it.

1. V-baseCollaborative workspaceOne real-time workspace for documents, code, spreadsheets, whiteboards, tasks, chat and meetings.TypeScript · Next.js · Convex · Liveblocks · LiveKit · FastAPI (vbase-rce)

   • SillyCatto/V-base on GitHub (opens in a new tab)
   • SillyCatto/vbase-rce on GitHub (opens in a new tab)

   The problem

   Teams split their work across separate apps for documents, code, spreadsheets, tasks, whiteboards, chat and calls, and lose context every time they switch.

   The idea

   One workspace with a room type for each of those jobs, all live: Liveblocks and Yjs sync edits and cursors, Convex stores everything that persists, LiveKit carries the calls, and Gemini helps inside code, document and whiteboard rooms.

   Outcome

   Deployed on Vercel. Code rooms can run code through the public Piston API or through vbase-rce, our own engine: a FastAPI service that runs each job in a locked-down Docker container with no network, a read-only filesystem, and memory, CPU and process limits.

2. BoibrittoLiterary community platformA home for readers and hobbyist writers: find books, track reading, discuss, and write.TypeScript · Next.js · Express.js · MongoDB · Firebase Auth

   • SillyCatto/boibritto on GitHub (opens in a new tab)

   The problem

   Readers and aspiring writers use separate platforms to discover books, track their reading, talk about them and share their own work.

   The idea

   One literary community: discovery through the Google Books API, personal collections, reading tracking, chapter-based writing, discussion forums with spoiler controls and Markdown blogs, each with its own visibility settings.

   Outcome

   A full-stack platform: a Next.js front end, an Express and MongoDB API, Google sign-in through Firebase, and an AdminJS panel for moderating content and users.

3. Pose2WordAI / ML researchWord-level sign language recognition, from raw video to a predicted word.Python · PyTorch · MediaPipe · OpenCV · Streamlit

   • SillyCatto/Pose2Word on GitHub (opens in a new tab)
   • SillyCatto/WLASL_video_keyframe_extractor on GitHub (opens in a new tab)

   The problem

   Recognising signed words reliably from raw, uneven video, to support communication tools for Deaf and Hard-of-Hearing people.

   The idea

   Clean each video (frame rate, contrast, signer crop, idle trimming), pick 8 to 15 keyframes from fused motion and landmark signals, reduce every frame to 56 MediaPipe landmarks, then classify the sequence with an LSTM or Transformer.

   Outcome

   A team project for our ML lab course: a Streamlit workbench that runs the whole pipeline, from upload to top-k prediction, over 15 words from WLASL. The keyframe step became its own tool, which combines MediaPipe landmarks with optical flow.

4. Classroom ProBrowser extensionFolders, pins and a compact sidebar for a crowded Google Classroom.JavaScript · HTML · CSS · Chrome Manifest V3 · SortableJS

   • SillyCatto/classroom-pro-extension on GitHub (opens in a new tab)

   The problem

   With many courses, Google Classroom gets crowded: important classes get buried and related ones can’t be grouped. Faculty with many course classrooms feel it most.

   The idea

   A Chrome side panel beside Classroom, rather than UI injected into its ever-changing page: folders, pins, nicknames, instant search and drag-and-drop ordering, with classes detected from the page automatically.

   Outcome

   A Manifest V3 extension in plain JavaScript, about 80 KB with no build step. Everything stays in the browser’s own extension storage: no account, no tracking, minimal permissions.

5. Signals for mCertiKOSOperating systemsPOSIX-style signals built into mCertiKOS, an educational x86 microkernel, for my OS course.C · x86 assembly · QEMU

   • SillyCatto/OS_Project-Signal on GitHub (opens in a new tab)

   The problem

   Without signals, a page fault in a user process, like a NULL dereference, ended in a kernel panic, and the shell couldn’t interrupt or kill a running program.

   The idea

   A pending-signal bitmask per process, checked on every return from a trap, with sigaction, kill and sigreturn system calls and a userspace trampoline that restores the process after its handler runs.

   Outcome

   SIGKILL ends a process unconditionally, SIGSEGV ends only the faulting process while the shell stays up, and Ctrl+C reaches the foreground process, which either dies or runs its own handler. The shell gained kill, trap, spawn and test commands.

6. AstraCppC++ libraryA minimal linear algebra library in C++, built for my SPL-1 course.C++17 · GoogleTest

   • SillyCatto/AstraCpp on GitHub (opens in a new tab)

   The problem

   Mathematical, engineering and scientific code keeps needing the same core linear algebra: vector and matrix operations, decompositions and solving systems of equations.

   The idea

   Keep it small and focused: four classes, Vector, Matrix, Decomposer and Solver, covering the fundamentals, including RREF, null spaces, LU and PLU decomposition, and a linear equation solver.

   Outcome

   Version 1.0, with a GoogleTest suite, built with two teammates. It runs on Windows, macOS and Linux, and every class is documented with examples on the project wiki.